Infant learning of phonological alternations is biased by phonetic similarity
Megha
b
Sundara ,
Yun Jung
a. University College London
Introduction
b
Kim
& Adam
b
Chong
b. UCLA
Study 1 (White & Sundara, 2014)
Question:
•  Infants are excellent at statistical learning:
•  Discriminating speech sounds (Anderson et al.,
Does 12-month-olds’ generalization of newly learned
alternations depend on phonetic similarity?
2003; Maye et al., 2002)
•  Phonotactics (Chambers et al., 2003)
•  Word segmentation (Saffran et al., 1996)
2008)
•  Evidence mounting that learners are biased by
phonetic similarity – they prefer alternations
between phonetically similar sounds.
•  Typology (Steriade, 2001; Hayes & J. White, sub.)
•  Adult artificial language studies (Skoruppa et al.,
2011; J. White, 2014)
•  Computational modeling (Peperkamp et al., 2006;
Wilson, 2006; J. White, 2013)
40 monolingual English-learning 12month-olds at UCLA.
Procedure:
•  Virtually no work with infant learners!!
Research questions:
1.  Are 12-month-olds biased by phonetic similarity
when generalizing newly learned alternations in
an artificial language? (Study 1)
2.  Are 12-month-olds biased by similarity when
learning alternations in their first language?
(Study 2)
b
Visual Fixation Procedure
rom poli
na poli
rom voli
na voli
rom timu
rom zimu
rom timu
na zimu
na timu
na zimu
rom timu
na zimu
Test pairs: buni/vuni, bagu/vagu, dilu/zilu, dari/zari
CONTROL condition – [b ~ v] or [d ~ z]
Labials Alternating
Coronals Alternating
rom boli
rom boli
na voli
na voli
rom boli
na boli
Exposure:
Should
not
generalize
≠
v
d
v
b
Should
not
generalize
z
p
Should
not
generalize
v
=
z
t
No difference in looking times
Results:
BIAS condition - [p ~ v] or [t ~ z]
Labials Alternating
Coronals Alternating
na voli
na voli
v
Difference in looking times
2. Test: Novel pairs of CVCV words (e.g. buni…vuni…)
rom poli
rom poli
p
Should
generalize
1.  Familiarization: Phrases providing evidence for an
alternation.
•  ‘Function’ element (na or rom) + CVCV
‘content’ word (e.g. rom poli…na timu…)
•  16 phrases, repeated twice per trial
•  3 trials (45s each, 135s total exposure)
Control condition
BIAS condition
Exposure:
Participants:
•  Tracking statistics likely plays a role in learning
phonological alternations as well.
•  Computational work (Peperkamp et al., 2006)
•  Experimental work with infants (K. White et al.,
Predictions (if biased by similarity):
rom voli
na voli
Alternating place
Contrastive place
12
Mean looking time (s)
James
a
White ,
*
10
8
6
4
2
0
BIAS condition
CONTROL condition
à  12-month-olds generalized the alternations to new
pairs of sounds that were more similar, but not to
ones that were less similar.
à  Similar results found with adult learners in a previous
study. (J. White, 2014)
rom dimu
rom zimu
rom dimu
na zimu
na dimu
na zimu
rom dimu
na zimu
Test pairs: puni/vuni, pagu/vagu, tilu/zilu, tari/zari
Study 2 (Sundara, Kim, White, & Chong, submitted)
Phonetic similarity:
Do 12-month-olds depend solely on input statistics when
learning phonological alternations in their first language,
or is this learning biased by phonetic similarity?
Tapping in American English:
/t/ and /d/ (partially) neutralized to [ɾ] in certain contexts:
/t/
/d/
/pæd/
[ˈpæɾɪŋ]
–ting/–ding words (tapping contexts only)
146
598
Procedure:
Headturn Preference Procedure (HPP)
•  Brent corpus (Brent & Siskind, 2001)
•  9 infant-mother dyads chosen (infant ages 0;9–2;2).
•  All words ending in –ting/–ding extracted.
44
Based solely on input statistics:
[t ~ ɾ] learned first.
Bias based on phonetic similarity: [d ~ ɾ] learned first.
48 monolingual English-learning 12-month-olds
(24 in each exp.)
Corpus search of infant-directed speech:
15
Predictions:
Participants:
/pæt/
[ɾ] / V__V[–stress]
•  [d] and [ɾ] more similar than [t] and [ɾ] (Herd et al., 2010)
–ting
–ding
1.  Familiarization: 2 passages (45s to each)
•  E.g. Cutting papers with scissors is a lot of fun…
Mommy is really good at cutting tofu…
•  Target word appeared 6 times/passage.
•  2 groups (cutting/meeting or patting/shooting,
counterbalanced)
2.  Test: 4 wordlists, 2 familiar and 2 novel (4 trials x 2
blocks).
Types
Tokens
•  Exp. 1: cut…cut…cut… (also: meet, pat, shoot)
à Infants hear far more –ting words than –ding words in IDS.
•  Exp. 2: cud…cud…cud… (also, meed, pad, shood)
General Discussion
•  Input statistics are not sufficient for explaining infants’ learning and generalization of
phonological alternations.
•  Instead, this learning is biased by phonetic similarity.
•  Previous experimental work with adults (J. White, 2014) and computational modeling (J.
White, 2013) suggests that this bias is a substantive bias favouring alternation between
perceptually similar sounds (i.e. a P-map bias, Steriade 2001).
•  Future infant work will look more into the type of bias (e.g. perceptual vs. features).
•  Prediction: alternations between similar sounds learned first, all else being equal.
Acknowledgements
We would like to thank Bruce Hayes, Sharon Peperkamp, and Kie Zuraw for helpful discussion; to Chad Vicenik,
Robyn Orfitelli, and Anya Mancillas for help in recruiting and testing infants; and to Jeffrey Lamontagne for help with
the corpus analysis in study 2. This research was funded by a COR Faculty Research Grant and NSF BCS-0951639 to
MS, and a University of Ottawa seed grant to JW.
Results:
14
Mean looking time (s)
Question:
*
12
10
Familiar
Novel
8
6
4
2
0
Exp. 1 [t ~ ɾ]
Exp. 2 [d ~ ɾ]
à 12-month-olds know [d ~ ɾ], but not [t ~ ɾ],
despite greater evidence for [t ~ ɾ] in the input.
à Exp. 3: Confirmed that 12-month-olds can
discriminate [d] and [ɾ] perceptually.
à  Note: if [t ~ ɾ] learned first, this would be a
saltatory alternation:
t
•
•
•
d
ɾ
Particularly egregious P-map violation.
Poses theoretical difficulties. (Hayes & J. White, sub.)
Disfavoured by learners (J. White, 2014).
References
Anderson, J. L., Morgan, J. L., & White, K. S. (2003). A statistical basis for speech sound discrimination. Language and Speech, 46(2–3), 155–182.
Brent, M. & Siskind, J. (2001). The role of exposure to isolated words in early vocabulary development. Cognition, 81(2), 31-44.
Chambers, K. E., Onishi, K. H., & Fisher, C. (2003). Infants learn phonotactic regularities from brief auditory experience. Cognition, 87(2), 69–77.
Hayes, B. & White, J. (submitted). Saltation and the P-map. Ms.
Herd, W., Jongman, A. & Sereno, J. (2010). An acoustic and perceptual analysis of /t/ and /d/ flaps in American English. Journal of Phonetics, 38(4),
504-516.
Maye, J., Werker, J F., & Gerken, L. (2002). Infant sensitivity to distributional information can affect phonetic discrimination. Cognition, 82(3), B101–
B111.
Peperkamp, S., Le Calvez, R., Nadal, J.-P., & Dupoux, E. (2006a). The acquisition of allophonic rules: Statistical learning with linguistic constraints.
Cognition, 101, B31–B41.
Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8-month-old infants. Science, 274, 1926–1928.
Skoruppa, K., Lambrechts, A., & Peperkamp, S. (2011). The role of phonetic distance in the acquisition of phonological alternations. In S. Lima, K.
Mullin, & B. Smith (Eds.), Proceedings of the 39th North Eastern Linguistics Conference (pp. 717–729). Somerville, MA: Cascadilla Press.
Steriade, D. (2001). The phonology of perceptibility effects: the P-map and its consequences for constraint organization. Ms.
Sundara, M., Kim, Y., White, J., & Chong, A. There is no pat in patting: Acquisition of phonological alternations by English-learning 12-month-olds. Ms.
White, J. (2013). Bias in Phonological Learning: Evidence from Saltation. Ph.D. dissertation, UCLA.
White, J. (2014). Evidence for a learning bias against saltatory phonological alternations. Cognition, 130(1), 96-115.
White, J. & Sundara, M. (2014). Biased generalization of newly learned phonological alternations by 12-month-old infants. Cognition, 133(1), 85–90.
White, K., Peperkamp, S., Kirk, C., & Morgan, J. (2008). Rapid acquisition of phonological alternations by infants. Cognition, 107, 238–265.
Wilson, C. (2006). Learning phonology with substantive bias: An experimental and computational study of velar palatalization. Cognitive Science, 30,
945–982.
Workshop on Learning Biases
in Natural and Artificial Language Acquisition
2014 Annual Meeting of the LAGB, Oxford